CN108182413B - Mine moving target detection and tracking identification method - Google Patents
Mine moving target detection and tracking identification method Download PDFInfo
- Publication number
- CN108182413B CN108182413B CN201711477567.8A CN201711477567A CN108182413B CN 108182413 B CN108182413 B CN 108182413B CN 201711477567 A CN201711477567 A CN 201711477567A CN 108182413 B CN108182413 B CN 108182413B
- Authority
- CN
- China
- Prior art keywords
- digital
- mine
- rcnn
- network
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Abstract
The invention discloses a mine moving target detection and tracking identification method, which comprises the following steps of firstly, making a training set, synthesizing a digital label of a mine moving target into an environment photo as an identity identifier by acquiring a mine environment photo, labeling the digital label in the synthesis photo, and taking the labeled synthesis photo as the training set; secondly, establishing an acceleration region convolutional neural network fast-RCNN, and training an acceleration region convolutional neural network fast-RCNN model which can be used for digital label detection by using a training set; thirdly, inputting the processed acquired video into a trained acceleration region convolutional neural network fast-RCNN model, and detecting a bounding box of the digital label region; fourthly, performing character segmentation operation on the digital label area in the boundary frame to obtain each digital character of the digital sequence, and sequentially transmitting the digital characters into a digital recognition model LeNet network for digital recognition; and fifthly, the LeNet network returns the identified digital sequence, and the identity information of the mine moving target is determined according to the identified mine moving target digital label. The invention can accurately and effectively detect, track and identify the mine moving target.
Description
Technical Field
The invention relates to the field of intelligent identification of machine vision and mine moving target detection, in particular to an acceleration region convolutional neural network mine moving target detection and tracking identification method based on machine vision.
Background
The intelligent mining is a development trend of safe and efficient intensive production of mines, researches on accurate detection, tracking and identification of moving targets such as underground operating personnel, locomotive equipment, working robots and the like are carried out, and the intelligent mining has important significance for guaranteeing danger avoidance of underground personnel, collision avoidance of vehicles and intelligent safe mining. The existing identification method for moving targets such as mine personnel, underground locomotives, mobile robots and the like mainly adopts a static radio frequency identification technology, but the method can only carry out the in-and-out identification of the underground moving targets through the mine RFID radio frequency identification technology, cannot realize the multi-dimensional information detection and tracking identification of the moving targets, has multipath interference especially in an underground NLOS environment, is greatly influenced by electromagnetic wave propagation attenuation and propagation delay, and is difficult to realize the real-time tracking and accurate identification of the mine moving targets. In recent years, a tracking and identifying technology based on machine vision has the characteristics of high identification precision, strong anti-interference capability, capability of obtaining target images remotely and the like, and becomes a research hotspot in the fields of intelligent monitoring, moving target detection and identification, visual navigation and the like. Therefore, in order to overcome the defects of the existing mine moving target identification technology, the invention provides an acceleration convolution neural network method based on machine vision, and accurate detection, tracking and identification of the mine moving target are realized.
Disclosure of Invention
The technical scheme adopted by the invention is as follows: aiming at the existing problems, a mine moving target detection and tracking identification method based on machine vision is provided, and the method is used for realizing the detection and tracking identification of mine personnel, locomotives and robot moving targets. The mine moving target detection and tracking identification method comprises the following implementation steps:
step 1, making a training set: collecting mine environment photos by using mine video image collecting equipment, synthesizing the digital tags of mine moving targets serving as identity identifiers into the environment photos, labeling the digital tags in the synthetic photos, and using the labeled synthetic photos as training sets;
step 2, constructing and training an acceleration region convolutional neural network fast-RCNN: the accelerated region convolutional neural network Fast-RCNN comprises a convolutional layer part for extracting picture features, a full-connection layer part of a region suggestion network RPN and a full-connection layer part of a target detection network Fast-RCNN; the convolutional layer part for extracting the picture features uses a VGG16 network, the last classification layer output of the Fast-RCNN full connection layer part of the target detection network is set to be 2, and the regression layer output is set to be 8;
step 3, processing a video frame acquired by the mine video image acquisition equipment by using a trained acceleration region convolutional neural network fast-RCNN to acquire a digital tag region in the video frame;
step 4, performing character segmentation operation on the digital label obtained in the step 3 to obtain a single digital character;
step 5, inputting the single digital character obtained in the step 4 into a LeNet network of a digital recognition model for digital recognition;
and 6, determining the identity information of the mine moving target according to the digital label identified in the step 5.
And 7, if the digital label area is not detected in the video frame in the step 3, repeatedly executing the step 3 to the step 6.
The mine moving target detection and tracking identification method comprises the following steps of:
2.1) initialization of acceleration region convolutional neural network fast-RCNN: adopting an ImageNet pre-training model VGG16 to initialize a convolution layer part of a region suggestion network RPN, adopting a Gaussian distribution initialization region with zero mean standard deviation of 0.01 to suggest a full-connection layer part behind a network RPN convolution layer, and adopting an ImageNet pre-training model VGG16 to initialize a convolution layer part of a target detection network Fast-RCNN;
2.2) utilizing the training set in the step 1 to carry out end-to-end training on the RPN until the RPN converges;
2.3) generating a suggestion box by using the converged region suggestion network RPN, and using the suggestion box as input for training a target detection network Fast-RCNN alone and for fine-tuning an ImageNet pre-training model VGG16 network;
2.4) fixing parameters of a convolution layer part of the target detection network Fast-RCNN, and training a region suggestion network RPN by using the trained target detection network Fast-RCNN in the step 2.3);
2.5) fixing parameters of a convolution layer part of the target detection network Fast-RCNN, generating an advice frame according to the regional advice network RPN trained in the step 2.4), and training parameters of a full connection layer part of the target detection network Fast-RCNN;
2.6) repeating step 2.4) and step 2.5) until convergence of the fast-RCNN of the acceleration region convolutional neural network.
In the method for detecting, tracking and identifying the mine moving target, the step 1 of labeling the digital tags refers to framing the digital tags in the synthetic photos by using a rectangular frame, and recording the number of the digital tags in each synthetic photo and four-dimensional coordinate information of each digital tag boundary frame.
The mine moving target detection and tracking identification method is characterized in that the mine video image acquisition equipment comprises a mine intrinsic safety type visual sensor, a mine intrinsic safety type camera and a mine intrinsic safety type video camera.
The invention has the beneficial effects that:
compared with the existing mine personnel target identification method based on wireless radio frequency technologies such as WSN, RSSI and RFID, the method has the advantages of strong anti-interference capability, high identification precision and the like, can realize accurate detection and real-time tracking identification of targets such as underground personnel, locomotive equipment and mobile robots by adopting an acceleration region convolutional neural network method based on machine vision, and has important significance for improving intelligent monitoring of mines, tracking and identifying underground mobile targets in real time, and ensuring danger avoidance of underground personnel, collision avoidance of vehicles and intelligent safe mining.
Drawings
FIG. 1 is a flow chart of a mine moving target detection and tracking identification method based on machine vision
FIG. 2 is a flow chart of sample training set generation
FIG. 3 is a flow chart of the method for constructing and training the acceleration region convolutional neural network fast-RCNN model
FIG. 4 is a flow chart of character segmentation
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention are described in detail below with reference to the accompanying drawings.
Fig. 1 is a flow chart of a mine moving target detection and tracking identification method based on machine vision. The realization process is as follows: the method comprises the steps of collecting video frame pictures from mine video image collecting equipment, making the video frame pictures into pictures with proper formats, inputting the pictures into a trained acceleration region convolutional neural network fast-RCNN detection model, carrying out character segmentation operation on digital labels when the digital labels in the video frame pictures are detected, sequentially inputting the segmented single digital characters into a LeNet network for digital recognition, finally returning recognized digital sequences, and determining identity information of a mine moving target according to the recognized digital sequences.
Fig. 2 is a flow chart of sample training set generation. Collecting a mine multi-scene multi-angle environment photo through mine video image collecting equipment, synthesizing a digital label of a mine moving target into the environment photo as an identity identifier, wherein the position of the digital label in the synthesized photo is random and can be placed in parallel to the synthesized photo or perpendicular to the synthesized photo; the position labeling of the digital label in the synthetic photo is carried out, and the labeling of the digital label refers to that the digital label in the synthetic photo is framed by a rectangular frame, and the number of the digital labels in each synthetic photo and the four-dimensional coordinate information of each digital label boundary frame are recorded; and carrying out format conversion on the marked synthetic photo to generate a training set.
FIG. 3 is a flow chart of sample set generation for training the region convolutional neural network Faster-RCNN model. Specifically, the method for constructing and training the acceleration region convolutional neural network fast-RCNN model flow chart comprises the following steps:
step 1, initializing a convolution layer part of a region suggestion network RPN by using an ImageNet pre-training model VGG16 network, and initializing a full-connection layer part behind the region suggestion network RPN convolution layer by using Gaussian distribution with zero mean standard deviation of 0.01;
step 2, inputting the training set into the initialized convolutional layer part of the region recommendation network RPN in the step 1, generating a feature plane, training the region recommendation network RPN by using the feature plane until the region recommendation network RPN is converged, wherein a loss function required for training the region recommendation network RPN is as follows:
where i is the index of the candidate region in a training batch, piIs the prediction probability that the ith candidate region is the target. GT label if the candidate area is positiveThen is 1, otherwiseIs 0. t is tiIs a vector, i.e. ti=(tx,ty,tw,th) 4 parametric coordinates representing the predicted bounding box,is a coordinate vector of the GT bounding box corresponding to the positive candidate region, i.e.λ is the balance weight, here taken as 10, NclsTo train batch size, i.e. 256, NregThe number of candidate regions is about 2000. Loss of classification LclsIs the log loss of two classes (target and non-target), namely:
further comprises
Solving and calculating regression loss Lreg。
Wherein, R is a loss function with robustness.Meaning that only positive candidate regions, i.e.There is a regression loss at all times, in other cases, i.e.There was no regression loss.
Wherein x, y, w and h are respectively the central coordinate (x, y), width and height of the prediction bounding box; x is the number ofa、ya、wa、haRespectively, the center coordinates (x) of the bounding box of the candidate regiona,ya) Width and height; x is the number of*、y*、w*、h*Respectively, the center coordinates (x) of the GT bounding box*,y*) Width and height. t is tiAndfor calculating the regression loss, we understand the boundary regression from the candidate region bounding box to the nearby GT bounding box.
Step 3, initializing a convolutional layer part of a target detection network Fast-RCNN by using an ImageNet pre-training model VGG16 network, inputting a training set into the convolutional layer part of the target detection network Fast-RCNN, and generating a characteristic surface;
step 4, independently training a target detection network Fast-RCNN by using a suggestion frame generated by the region suggestion network RPN converged in the step 2 and the feature plane generated in the step 3 as input, so that the ImageNet pre-training model VGG16 network mentioned in the step 3 is finely adjusted;
step 5, fixing the parameters of the convolutional layer part of the target detection network Fast-RCNN mentioned in the step 3, and training a region suggestion network RPN by using the trained target detection network Fast-RCNN;
step 6, fixing the parameters of the convolutional layer part of the target detection network Fast-RCNN mentioned in the step 3, and using the region suggestion network RPN trained in the step 5 to generate a suggestion frame to train the parameters of the full connection layer part of the Fast-RCNN;
and 7, repeating the step 5 and the step 6 until the acceleration region convolutional neural network fast-RCNN converges.
FIG. 4 is a flow chart of character segmentation. And carrying out binarization processing on the acquired digital label, acquiring the outline of each digit in the digital label by using an outline-taking method, intercepting a rectangular picture along the outline of each digit, and sequentially inputting the rectangular picture into a digital identification network for digital identification.
Obviously, it should be understood by those skilled in the art that the identification method according to the present invention and the above embodiments is not only applied to the coal mine underground environment as the mine moving target monitoring and tracking identification, but also applied to the monitoring, tracking and positioning of moving targets in non-coal mines such as metal and nonmetal, and the tracking identification and intelligent visual monitoring of the moving operation equipment on the underground intelligent working face. The invention does not limit the communication technical fields of non-coal mines, intelligent working face mobile monitoring, accurate identification and positioning of Internet of things equipment and the like except for underground mobile target positioning of coal mines.
While the invention has been described in detail with reference to specific preferred embodiments thereof, it will be understood by those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (3)
1. A mine moving target detection and tracking identification method is characterized in that a machine vision-based acceleration region convolutional neural network algorithm is adopted for realizing target detection and tracking identification of mine personnel, locomotives and mobile robots, and comprises the following steps:
step 1, making a training set: collecting mine environment photos by using mine video image collecting equipment, synthesizing the digital tags of mine moving targets serving as identity identifiers into the environment photos, labeling the digital tags in the synthetic photos, and using the labeled synthetic photos as training sets;
step 2, constructing and training an acceleration region convolutional neural network fast-RCNN: the accelerated region convolutional neural network Fast-RCNN comprises a convolutional layer part for extracting picture features, a full-connection layer part of a region suggestion network RPN and a full-connection layer part of a target detection network Fast-RCNN; the convolutional layer part for extracting the picture features uses a VGG16 network, the last classification layer output of the Fast-RCNN full connection layer part of the target detection network is set to be 2, and the regression layer output is set to be 8;
step 3, processing a video frame acquired by the mine video image acquisition equipment by using a trained acceleration region convolutional neural network fast-RCNN to acquire a digital tag region in the video frame;
step 4, performing character segmentation operation on the digital label obtained in the step 3 to obtain a single digital character;
step 5, inputting the single digital character obtained in the step 4 into a LeNet network of a digital recognition model for digital recognition;
step 6, determining the identity information of the mine moving target according to the digital label identified in the step 5;
7, if the digital label area is not detected in the video frame in the step 3, repeatedly executing the step 3 to the step 6;
it is also characterized in that step 2 further comprises the following substeps:
2.1) initialization of acceleration region convolutional neural network fast-RCNN: adopting an ImageNet pre-training model VGG16 to initialize a convolution layer part of a region suggestion network RPN, adopting a Gaussian distribution initialization region with zero mean standard deviation of 0.01 to suggest a full-connection layer part behind a network RPN convolution layer, and adopting an ImageNet pre-training model VGG16 to initialize a convolution layer part of a target detection network Fast-RCNN;
2.2) utilizing the training set in the step 1 to carry out end-to-end training on the RPN until the RPN converges;
2.3) generating a suggestion box by using the converged region suggestion network RPN, and using the suggestion box as input for training a target detection network Fast-RCNN alone and for fine-tuning an ImageNet pre-training model VGG16 network;
2.4) fixing parameters of a convolution layer part of the target detection network Fast-RCNN, and training a region suggestion network RPN by using the trained target detection network Fast-RCNN in the step 2.3);
2.5) fixing parameters of a convolution layer part of the target detection network Fast-RCNN, generating an advice frame according to the regional advice network RPN trained in the step 2.4), and training parameters of a full connection layer part of the target detection network Fast-RCNN;
2.6) repeating step 2.4) and step 2.5) until convergence of the fast-RCNN of the acceleration region convolutional neural network.
2. The method for detecting, tracking and identifying the mine moving object according to claim 1, wherein the step 1 of labeling the digital tags means framing the digital tags in the composite photo with a rectangular frame, and recording the number of the digital tags in each composite photo and four-dimensional coordinate information of each digital tag bounding box.
3. The mine moving object detecting, tracking and identifying method according to claim 1, wherein the mine video image acquisition equipment comprises a mine intrinsic safety type vision sensor, a mine intrinsic safety type camera and a mine intrinsic safety type video camera.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711477567.8A CN108182413B (en) | 2017-12-29 | 2017-12-29 | Mine moving target detection and tracking identification method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711477567.8A CN108182413B (en) | 2017-12-29 | 2017-12-29 | Mine moving target detection and tracking identification method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108182413A CN108182413A (en) | 2018-06-19 |
CN108182413B true CN108182413B (en) | 2022-01-25 |
Family
ID=62548958
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711477567.8A Active CN108182413B (en) | 2017-12-29 | 2017-12-29 | Mine moving target detection and tracking identification method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108182413B (en) |
Families Citing this family (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109685066B (en) * | 2018-12-24 | 2021-03-09 | 中国矿业大学(北京) | Mine target detection and identification method based on deep convolutional neural network |
CN109977853B (en) * | 2019-03-25 | 2023-07-14 | 太原理工大学 | Underground worker panoramic monitoring method based on multiple identification devices |
CN109977958A (en) * | 2019-03-25 | 2019-07-05 | 中国科学技术大学 | A kind of offline handwritten form mathematical formulae identification reconstructing method |
CN110009628A (en) * | 2019-04-12 | 2019-07-12 | 南京大学 | A kind of automatic testing method for polymorphic target in continuous two dimensional image |
CN110633641A (en) * | 2019-08-15 | 2019-12-31 | 河北工业大学 | Intelligent security pedestrian detection method, system and device and storage medium |
CN110569843B (en) * | 2019-09-09 | 2022-02-08 | 中国矿业大学(北京) | Intelligent detection and identification method for mine target |
CN110909675A (en) * | 2019-11-22 | 2020-03-24 | 广州供电局有限公司 | Method and device for identifying violation behaviors, computer equipment and storage medium |
CN110991385A (en) * | 2019-12-13 | 2020-04-10 | 珠海大横琴科技发展有限公司 | Method and device for identifying ship driving track and electronic equipment |
CN111539422B (en) * | 2020-04-13 | 2022-09-06 | 南京理工大学 | Flight target cooperative identification method based on fast RCNN |
WO2021211068A1 (en) * | 2020-04-15 | 2021-10-21 | Aselsan Elektroni̇k Sanayi̇ Ve Ti̇caret Anoni̇m Şi̇rketi̇ | A method for training shallow convolutional neural networks for infrared target detection using a two-phase learning strategy |
CN111428813A (en) * | 2020-04-16 | 2020-07-17 | 西安近代化学研究所 | Panel number identification and pressing method based on deep learning |
CN111582339B (en) * | 2020-04-28 | 2023-07-25 | 江西理工大学 | Vehicle detection and recognition method based on deep learning |
CN111968102B (en) * | 2020-08-27 | 2023-04-07 | 中冶赛迪信息技术(重庆)有限公司 | Target equipment detection method, system, medium and electronic terminal |
CN112633287B (en) * | 2020-12-25 | 2024-04-26 | 中国矿业大学 | Text recognition method and device for mine multi-source heterogeneous graphic information |
CN116821651A (en) * | 2023-05-04 | 2023-09-29 | 广州海洋地质调查局 | Method and device for setting natural gas hydrate distant view zone and finely identifying ore body |
CN117195138B (en) * | 2023-11-07 | 2024-02-20 | 湖南展通科技集团有限公司 | Production equipment safety production management method based on artificial intelligence and related device |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2016145547A1 (en) * | 2015-03-13 | 2016-09-22 | Xiaoou Tang | Apparatus and system for vehicle classification and verification |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106022232A (en) * | 2016-05-12 | 2016-10-12 | 成都新舟锐视科技有限公司 | License plate detection method based on deep learning |
CN106780612B (en) * | 2016-12-29 | 2019-09-17 | 浙江大华技术股份有限公司 | Object detecting method and device in a kind of image |
CN106845430A (en) * | 2017-02-06 | 2017-06-13 | 东华大学 | Pedestrian detection and tracking based on acceleration region convolutional neural networks |
CN107016357B (en) * | 2017-03-23 | 2020-06-16 | 北京工业大学 | Video pedestrian detection method based on time domain convolutional neural network |
CN107451607B (en) * | 2017-07-13 | 2018-09-18 | 山东中磁视讯股份有限公司 | A kind of personal identification method of the typical character based on deep learning |
-
2017
- 2017-12-29 CN CN201711477567.8A patent/CN108182413B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2016145547A1 (en) * | 2015-03-13 | 2016-09-22 | Xiaoou Tang | Apparatus and system for vehicle classification and verification |
Also Published As
Publication number | Publication date |
---|---|
CN108182413A (en) | 2018-06-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108182413B (en) | Mine moving target detection and tracking identification method | |
CN109685066B (en) | Mine target detection and identification method based on deep convolutional neural network | |
CN107907124B (en) | Positioning method based on scene recognition, electronic equipment, storage medium and system | |
CN106778712B (en) | Multi-target detection and tracking method | |
Elmezain et al. | A robust method for hand gesture segmentation and recognition using forward spotting scheme in conditional random fields | |
Rout | A survey on object detection and tracking algorithms | |
CN104615986B (en) | The method that pedestrian detection is carried out to the video image of scene changes using multi-detector | |
CN104077596A (en) | Landmark-free tracking registering method | |
CN103679698A (en) | System and method for detection and tracking of moving objects | |
CN105574848A (en) | A method and an apparatus for automatic segmentation of an object | |
CN113962274B (en) | Abnormity identification method and device, electronic equipment and storage medium | |
CN111784737B (en) | Automatic target tracking method and system based on unmanned aerial vehicle platform | |
CN108734109B (en) | Visual target tracking method and system for image sequence | |
CN103413149B (en) | Method for detecting and identifying static target in complicated background | |
CN106846367B (en) | A kind of Mobile object detection method of the complicated dynamic scene based on kinematic constraint optical flow method | |
CN104200492A (en) | Automatic detecting and tracking method for aerial video target based on trajectory constraint | |
CN113408584A (en) | RGB-D multi-modal feature fusion 3D target detection method | |
CN104299246A (en) | Production line object part motion detection and tracking method based on videos | |
CN112906777A (en) | Target detection method and device, electronic equipment and storage medium | |
Nayagam et al. | A survey on real time object detection and tracking algorithms | |
Fakhfakh et al. | Background subtraction and 3D localization of moving and stationary obstacles at level crossings | |
CN103679740A (en) | ROI (Region of Interest) extraction method of ground target of unmanned aerial vehicle | |
Yuan et al. | Fast QR code detection based on BING and AdaBoost-SVM | |
CN101833768A (en) | Method and system for carrying out reliability classification on motion vector in video | |
Shi et al. | TagAttention: Mobile object tracing without object appearance information by vision-RFID fusion |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |